LCF visits Higher School of Economics in St. Petersburg
12 June 2016
12 June 2016
Dr Nikolaos Kourentzes and several LCF members have visited the Higher School of Economics (HSE) in St. Petersburg end of May. The aim of the visit was to discuss possible scientific collaboration with HSE academics and take part in departmental research workshop for master students and academic staff.
The visit was organised by Professor Sergey Svetunkov, Head of Marketing Technologies master’s programme at HSE, who visited Lancaster Centre for Forecasting last autumn. During the workshop all LCF members presented recent research findings:
Nikolaos Kourentzes first delivered a talk on advances in promotional forecasting. The talk highlighted variety of impacts promotions can have on demand such as after promotion effects or product cannibalisation. Students learned how to take them into account in forecasting. Furthermore Nikolaos explained why multiplicative regressions are better suited to model sets of promotions on same products as seen often in practice. The talk ended with a case study of forecasting Greek economy with a text mining approach. It highlighted the importance of taking qualitative indicators into account.
The second talk was given by Yves Sagaert from Ghent University, visiting researcher at LCF on The value of external information: including leading indicators in sales forecasting. Yves presented recent findings on selecting variables with LASSO. The case is based on a car tyre supplier which uses several thousands of macroeconomic indicators in order to forecast demand with limited numbers of sales observations. Using shrinkage methods allows to overcome overfitting problem when dealing with such a large amount of indicators on short samples. Results from this experiment have highlighted 10 potential indicators which allowed forecasting model to outperform several univariate benchmark models.
After that Oliver Schaer presented recent findings on using search traffic information for pre-launch forecasting. This fully automated analogy based approach tries to address the difficulties companies have in finding the future market size of their new product generation. Including information from Google Trends on a video games sales dataset showed improvements in initial experiments over solely analogy based information.
The last talk was given by Ivan Svetunkov, research associate of LCF on topic of decreasing uncertainty in decision making using trace forecast likelihood. He presented results of recent research in estimation methods of univariate models. It was shown in the presentation that existing estimators based on multiple steps ahead forecast errors impose shrinkage on parameters of models. This can be beneficial in some cases but at the same time may cause problems if parameters over shrink. Ivan proposed a new estimator based on multivariate likelihood function that imposes a weakened shrinkage on parameters making them consistent, more efficient and unbiased.
The talks involved active discussions with the audience and both HSE and LCF representatives found the workshop to be very stimulating and further strengthened the scientific collaboration.